Skip to content

fix(bench): stop hardcoding the run date and replace GNU-only grep -oP in run_bench.sh #1899

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
inureyes merged 1 commit into from
Sep 24, 2026
+24 −11
Merged

fix(bench): stop hardcoding the run date and replace GNU-only grep -oP in run_bench.sh #1899

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
35 changes: 24 additions & 11 deletions benchmarks/run_bench.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,29 +1,42 @@
#!/bin/bash
# Single-model benchmark runner
# Usage: ./run_bench.sh <model_path> <csv_file>
set -euo pipefail

if [ $# -lt 2 ]; then
echo "Usage: $0 <model_path> <csv_file>" >&2
exit 1
fi

MODEL_PATH="$1"
CSV_FILE="$2"
MODEL_NAME=$(basename "$MODEL_PATH")
PROMPT="Explain the concept of machine learning in simple terms."
MAX_TOKENS=100
DATE="2026-03-15"
HARDWARE="NVIDIA_GB10_CUDA13.0"
MLX_VERSION="0.31.1"
DATE=$(date +%F)
HARDWARE="${HARDWARE:-NVIDIA_GB10_CUDA13.0}"
MLX_VERSION="${MLX_VERSION:-0.31.1}"
BUILD_TYPE="release"
BINARY="./target/release/mlxcel"

echo ">>> Benchmarking: $MODEL_NAME"

OUTPUT=$($BINARY generate -m "$MODEL_PATH" -p "$PROMPT" -n $MAX_TOKENS --profile 2>&1)
# A failed run is a row too -- let it reach the FAILED branch instead of
# aborting under `set -e`.
OUTPUT=$($BINARY generate -m "$MODEL_PATH" -p "$PROMPT" -n $MAX_TOKENS --profile 2>&1) || true

# Parse results. `sed -n ...p` rather than grep: a run that produced no stats
# must still reach the FAILED row below, and grep's no-match exit 1 would
# abort it under `set -e`.
field() { printf '%s\n' "$OUTPUT" | sed -n -E "s/^[[:space:]]*$1:[[:space:]]*([0-9.]+).*/\1/p"; }
tok_per_sec() { printf '%s\n' "$OUTPUT" | sed -n -E "s/^[[:space:]]*$1:.*\(([0-9.]+) tok\/s\).*/\1/p"; }

# Parse results
PROMPT_TOKENS=$(echo "$OUTPUT" | grep "Prompt tokens:" | sed 's/.*: *//')
GEN_TOKENS=$(echo "$OUTPUT" | grep "Generated tokens:" | sed 's/.*: *//')
PREFILL_MS=$(echo "$OUTPUT" | grep "Prefill:" | sed 's/.*: *//' | sed 's/ ms.*//')
PREFILL_TOKS=$(echo "$OUTPUT" | grep "Prefill:" | grep -oP '[\d.]+(?= tok/s)')
DECODE_MS=$(echo "$OUTPUT" | grep "Decode:" | sed 's/.*: *//' | sed 's/ ms.*//')
DECODE_TOKS=$(echo "$OUTPUT" | grep "Decode:" | grep -oP '[\d.]+(?= tok/s)')
PROMPT_TOKENS=$(field "Prompt tokens")
GEN_TOKENS=$(field "Generated tokens")
PREFILL_MS=$(field "Prefill")
PREFILL_TOKS=$(tok_per_sec "Prefill")
DECODE_MS=$(field "Decode")
DECODE_TOKS=$(tok_per_sec "Decode")

if [ -z "$DECODE_TOKS" ]; then
echo " FAILED or no output"
Expand Down
Loading